Method for estimating road surface adhesion coefficient considering transient characteristics of tire under lateral working conditions

Abstract

The present invention provides a method for estimating the road adhesion coefficient considering the transient characteristics of tires under lateral conditions, including the following steps: Step 1: Vehicle dynamics modeling; Step 2: Estimation of the road adhesion coefficient based on the unscented Kalman filter; Step 3: Adjustment of the UKF estimation error; Step 4: Time delay estimation. The present invention solves the problem of inaccurate estimation of the road adhesion coefficient caused by the lateral response lag generated by the transient characteristics of tires; at the same time, the present invention uses the time delay estimation method to perform time series delay estimation on the estimated value and the measured value of the vehicle lateral acceleration, and corrects the lateral slip ratio on the steady-state tire model, so that the time sequence between the estimated value and the measured value of the lateral acceleration is consistent, thereby improving the recognition accuracy of the road adhesion coefficient.

Description

Technical Field

[0001] The present invention relates to the technical field of vehicle dynamics model parameter estimation. Background Art

[0002] The magnitude of the road surface adhesion coefficient directly determines the motion limit boundary of the vehicle. That is, when the road surface adhesion coefficient is low, the force that the ground can provide for the vehicle motion is small, and compared with the high road surface adhesion coefficient, the vehicle tire force is more likely to saturate. Therefore, the road surface adhesion coefficient is a key variable in the vehicle chassis motion control system. Existing road surface adhesion coefficient identification methods usually rely on a Kalman filter estimator to identify the current road surface adhesion coefficient using vehicle state variables and control variables. Due to the limitation of the continuous excitation condition, these methods impose strict restrictions on the identification working conditions. For example, the road surface adhesion coefficient can only be identified when the vehicle is driving in a straight line with an acceleration greater than a threshold. This limitation greatly restricts the accuracy and real-time performance of the road surface adhesion coefficient estimation method.

[0003] Compared with the road surface adhesion coefficient identification based on longitudinal working conditions, the real-time estimation of the coefficient under lateral working conditions is more important for the chassis control system. The vehicle yaw motion limit is calculated in real time according to the estimated coefficient to prevent the vehicle from becoming unstable under lateral motion working conditions. However, existing road surface adhesion coefficient estimation methods based on a Kalman filter estimator all adopt a steady-state tire model, ignoring the transient characteristics of the tire during lateral motion, resulting in a phase lag between the estimated value based on the steady-state model and the measured value, which directly affects the accuracy of the road surface adhesion coefficient estimation. Summary of the Invention

[0004] In order to ensure the estimation accuracy of the road surface adhesion coefficient during vehicle lateral motion, the present invention proposes a road surface adhesion coefficient estimation method considering tire transient characteristics under lateral working conditions.

[0005] The technical solution of the present invention is as follows:

[0006] A road surface adhesion coefficient estimation method considering tire transient characteristics under lateral working conditions includes the following steps:

[0007] Step 1: Vehicle dynamics modeling

[0008] For the problem of road surface adhesion coefficient estimation under lateral working conditions, a three-degree-of-freedom vehicle model is selected. The three-degree-of-freedom vehicle model includes three degrees of freedom: longitudinal motion, lateral motion, and rotation about the z-axis of the vehicle coordinate system;

[0009] Step 2: Road surface adhesion coefficient estimation based on unscented Kalman filter

[0010] Based on the vehicle dynamics model established in Step 1, the relationship between the road surface adhesion coefficient and the observed variables (longitudinal acceleration, lateral acceleration, and yaw rate) was obtained. Taking the road surface adhesion coefficient as the state variable and the observed variables as In the vehicle dynamics model, the vehicle state variables (longitudinal speed, lateral speed, front and rear axle wheel speeds, steering wheel angle, etc.) are time-varying parameters, and the vehicle structure parameters (distance from the center of mass to the front axle, wheelbase, track width, steering ratio, and effective tire radius, etc.) are fixed parameters;

[0011] Step 3: UKF Estimation Error Adjustment

[0012] When the slip rate is small, the relationship between the tire force and the slip rate is consistent under different road surface adhesion coefficients, and the original value is maintained. When the estimation error of the lateral acceleration is large, the P covariance matrix is updated;

[0013] When it is detected that the error between the estimated value and the measured value of a y is greater than ε, the covariance matrix P will be reset to P0,

[0014]

[0015] where a y is the measured value of the sensor, is the current estimated value of UKF, and ε is the parameter to be calibrated

[0016] Step 4: Time Delay Estimation

[0017] The time delay between the measured value and the estimated value of the lateral acceleration is calculated through the time cross-correlation function, and this delay time is added to the estimated value of the lateral acceleration, thereby realizing the timing calibration between the measured value and the estimated value.

[0018] The method of the present invention solves the problem of inaccurate estimation of the road surface adhesion coefficient caused by the lateral response lag generated by the transient characteristics of the tire; the time series delay estimation is performed on the estimated value and the measured value of the vehicle lateral acceleration by using the time delay estimation method, and the lateral slip rate is corrected on the steady-state tire model, so that the timing between the estimated value and the measured value of the lateral acceleration is consistent, thereby improving the recognition accuracy of the road surface adhesion coefficient. Description of the Drawings

[0019] Figure 1 is the block diagram of the road surface adhesion coefficient estimation method considering the transient characteristics of the tire under the lateral condition;

[0020] Figure 2 is the relationship between the slip rate and the nominal tire force;

[0021] Figure 3 is the recognition result of the road surface adhesion coefficient;

[0022] Figure 4 It is a graph showing the variation of the estimated value and the actual value of the lateral acceleration;

[0023] Figure 5 It is a schematic diagram of the vehicle model;

[0024] Figure 6 It is a simulation comparison graph of the road surface adhesion coefficient;

[0025] Figure 7 It is a simulation comparison graph of the lateral acceleration;

[0026] Figure 8 It is a graph showing the change of the front wheel steering angle during the simulation;

[0027] Figure 9 It is a graph showing the change of the vehicle speed during the simulation. Specific implementation manners

[0028] The technical solution provided by the present application will be further described below in conjunction with specific embodiments and their accompanying drawings. In combination with the following description, the advantages and features of the present application will be more clearly understood.

[0029] The present invention takes into account the transient characteristics of the tire and proposes a road surface adhesion coefficient identification method that integrates delay estimation, and conducts simulation verification in the well-known dynamic simulation software Carsim. The simulation results under the standard double lane change condition show that the method proposed by the present invention has a great improvement in multiple aspects such as convergence time, identification accuracy, and static error compared with the comparative method.

[0030] The structural block diagram of the road surface adhesion coefficient estimation method considering the transient characteristics of the tire under the lateral condition of the present invention is as Figure 1 shown, and the process is as follows:

[0031] Step 1: Vehicle dynamics modeling

[0032] For the problem of road surface adhesion coefficient estimation under the lateral condition, the present invention selects a three-degree-of-freedom vehicle model. This three-degree-of-freedom vehicle model includes three degrees of freedom: longitudinal motion, lateral motion, and rotation about the z-axis of the vehicle coordinate system. The interaction between the road surface and the tire directly determines the vehicle motion characteristics. Therefore, selecting a suitable tire model is the key to constructing the vehicle dynamics model. Since the Burckhardt tire model can describe the tire force coupling characteristics with a simple structure, the tire force of this model is expressed in exponential form with a simple structure, which is suitable for a large number of calculations of the present invention. If a more accurate tire model is used here, the effect will be consistent with our analysis, but more tire model parameters are involved, and at the same time the expression is more complex. At the same time, a more accurate high-complexity tire model will greatly increase the difficulty of identifying the road surface adhesion coefficient.

[0033] Step 2: Estimation of road surface adhesion coefficient based on unscented Kalman filter

[0034] Based on the vehicle dynamics model established in Step 1, the relationship between the road surface adhesion coefficient and the observed variables (longitudinal acceleration, lateral acceleration, and yaw rate) was obtained. Considering that the above vehicle dynamics model is a nonlinear model, we used the unscented Kalman filter method to construct a basic method for identifying the road surface adhesion coefficient. Taking the road surface adhesion coefficient as the state variable and the observed variables as The vehicle state variables (longitudinal speed, lateral speed, front and rear axle wheel speeds, steering wheel angle, etc.) in the vehicle dynamics model are time-varying parameters, and the vehicle structure parameters (distance from the center of mass to the front axle, wheelbase, track width, steering ratio, and effective tire radius, etc.) are fixed parameters.

[0035] Step 3: UKF Estimation Error Adjustment

[0036] For time-varying systems, traditional unscented Kalman filtering can achieve high estimation accuracy. However, as the unscented Kalman filtering process progresses, the covariance matrix is continuously iterated, and the influence of the accumulation of old data on the covariance matrix P gradually increases, causing the covariance matrix P to gradually become smaller, and then making the estimation process tend to be stable. This results in the inability of the unscented Kalman filter to quickly and accurately track parameter changes. The road surface adhesion coefficient is jointly determined by the tire and road surface types and is greatly affected by the type of driving road. When the road surface type undergoes a sudden change, such as from a tarmac road to a snow-covered road, the road surface adhesion coefficient will also change suddenly with the road surface type. At this time, new data can better reflect the parameter change situation than old data.

[0037] The variation relationship between the nominal tire force f(s i ) and under different road surface conditions, as Figure 2 shown.

[0038] It can be found from Figure 2 that when the slip ratio is small, that is, the four-wheel tire force is small, the relationship between the tire force and the slip ratio is almost the same under different road surface adhesion coefficients. As a result, under the conditions of vehicle uniform motion and small acceleration, the slip ratio and tire force of the tire under different road surface adhesion coefficients are almost the same. At this time, it is impossible to identify the road surface adhesion coefficient only based on the vehicle chassis information. Therefore, we choose to maintain the original value and update the P covariance matrix when the lateral acceleration estimation error is large, so as to quickly identify the road surface adhesion coefficient value.

[0039] Therefore, when it is detected that the error between the estimated value and the measured value of a y is greater than ε, the covariance matrix P will be reset.

[0040]

[0041] Among them, a y is the measured value of the sensor, is the current estimated value of UKF, and ε is the parameter to be calibrated

[0042] Step 4: Delay estimation

[0043] It can be found from Step 3 that the judgment condition for resetting the covariance matrix is the difference between the measured value and the model estimated value of the lateral acceleration. The change process of the lateral acceleration is similar to a sine signal, with positive and negative values alternating, and the range of change of the acceleration derivative is relatively large. When the acceleration derivative is large, the time delay of the lateral movement of the tire will cause a large error, which will in turn cause the covariance matrix to be mis-triggered. Moreover, when the acceleration is large, it is often at the positive and negative conversion points of the lateral acceleration. At this time, the tire sideslip angle is near 0, which just does not meet the excitation condition for identifying the road adhesion coefficient. If the covariance matrix is reset at this time, the road adhesion coefficient will oscillate violently as Figure 3 shown, the estimated value of the road adhesion coefficient oscillates when the lateral acceleration alternates between positive and negative, and the steady-state value of the road adhesion coefficient is very different from the true value.

[0044] The present invention calculates the time delay between the measured value and the estimated value of the lateral acceleration through the time cross-correlation function, and adds this delay time to the estimated value of the lateral acceleration, thereby realizing the timing calibration between the measured value and the estimated value.

[0045] Step 5: Software-in-the-loop simulation verification

[0046] According to the method proposed above, verification was carried out in the carsim-simulink co-simulation environment.

[0047] The above major steps are described in detail as follows:

[0048] Step 1: Vehicle model modeling

[0049] For the problem of identifying the road adhesion coefficient under lateral working conditions, the present invention selects the vehicle three-degree-of-freedom model as shown below

[0050] ma x =F xr -F yf sinδ+F xf cosδ

[0051] ma y =E yr +F yf cosδ+F xf sinδ

[0052]

[0053] The schematic diagram of this vehicle dynamics model is as Figure 5 shown, where m is the vehicle mass; Fxf , F xr are the longitudinal forces of the front and rear axles of the vehicle respectively; F yf , F yr are the lateral forces of the front and rear axles of the vehicle respectively; a x , a y are the longitudinal acceleration, lateral acceleration and yaw angular velocity of the vehicle respectively; I z is the moment of inertia of the vehicle about the Z-axis; a and b are the distances from the center of mass to the front axle and the rear axle respectively; δ is the steering angle of the front wheels of the vehicle.

[0054] Then, the formulas for the longitudinal and lateral forces of the front and rear axles in the above three-degree-of-freedom vehicle model are established. The present invention adopts the Burckhardt model, and the formulas are as follows:

[0055]

[0056]

[0057]

[0058]

[0059]

[0060]

[0061] Among them, s xi is the longitudinal slip ratio of the front and rear axles, i = {f, r}; r dyn is the effective radius of the tire; w wi is the wheel speed of the front and rear axles; v x is the longitudinal speed at the center of mass of the vehicle; v y is the lateral speed at the center of mass of the vehicle; is the yaw angular velocity at the center of mass of the vehicle; s i is the combined slip ratio of the vehicle; f(s i ) is the nominal tire force; s yf is the lateral slip ratio of the front axle; s yr is the lateral slip ratio of the rear axle; F zi is the vertical load before and after the vehicle; c1 and c2 are the parameters in the tire model, and these parameters are determined by the tire characteristics; μ is the road adhesion coefficient, which is determined by the vehicle driving environment.

[0062]

[0063]

[0064] Among them, L is the distance between the front and rear axles of the vehicle; g is the acceleration due to gravity; h is the height of the center of mass from the ground.

[0065] Step 2: Estimation of road surface adhesion coefficient based on unscented Kalman filter

[0066] Based on the vehicle dynamics model established in Step 1, the relationship between the road surface adhesion coefficient and the observed variables (longitudinal acceleration, lateral acceleration, and yaw rate) is obtained. Considering that the above vehicle dynamics model is a non-linear model, an unscented Kalman filter method is used to construct a basic method for identifying the road surface adhesion coefficient. Taking the road surface adhesion coefficient as the state variable and the observed variables as The vehicle state variables (longitudinal speed, lateral speed, four-wheel speeds, and steering wheel angle, etc.) in the vehicle dynamics model are time-varying parameters, and the vehicle structure parameters (distance from the center of mass to the front axle, wheelbase, track width, steering ratio, and effective tire radius, etc.) are fixed parameters.

[0067] The standard UKF process is prior art and will not be described in the specification of the present invention. In the unscented Kalman filter (UKF) process, the state quantity and covariance P are continuously updated to estimate the road surface adhesion coefficient.

[0068] Step 3: UKF estimation error adjustment

[0069] For a time-varying system, traditional unscented Kalman filter can achieve high estimation accuracy. However, as the unscented Kalman filter process progresses, the covariance matrix is continuously iterated, and the influence of the accumulation of old data on the covariance matrix P gradually increases, making the covariance matrix P gradually become smaller, and then making the estimation process tend to be stable, which causes the unscented Kalman filter to be unable to quickly and accurately track parameter changes. The road surface adhesion coefficient is jointly determined by the tire and the road surface type and is greatly affected by the type of the driving road. When the road surface type suddenly changes, such as from asphalt road to snow-covered road, the road surface adhesion coefficient will also suddenly change with the road surface type. At this time, new data can better reflect the parameter change situation than old data.

[0070] When the lateral acceleration estimation error is large, update the P covariance matrix to quickly identify the road surface adhesion coefficient value.

[0071] Therefore, when the estimated value of the lateral acceleration and the measured value a y have an error greater than ε, the covariance matrix P will be reset to P0 = 1×10 -5 .

[0072]

[0073] Among them, a y is the measured value of the sensor, is the current estimated value of UKF, and ε is the parameter to be calibrated

[0074] Step 4: Delay Estimation

[0075] It can be found from Step 3 that the judgment condition for resetting the covariance matrix is the difference between the measured value and the model estimated value of the lateral acceleration. The change process of the lateral acceleration is similar to a sine signal, with positive and negative values alternating, and the change range of the acceleration derivative is relatively large. When the acceleration derivative is large, the time delay of the lateral movement of the tire will cause a large error, which will in turn cause the covariance matrix to be triggered erroneously. Moreover, when the acceleration is large, it often occurs at the positive and negative conversion points of the lateral acceleration. At this time, the tire side slip angle is near 0, which just does not meet the excitation condition for road adhesion coefficient identification. If the covariance matrix is reset at this time, the road adhesion coefficient will oscillate violently, as Figure 3 shown. The estimated value of the road adhesion coefficient oscillates when the lateral acceleration alternates between positive and negative, and the steady-state value of the road adhesion coefficient differs greatly from the true value. The present invention uses the time cross-correlation function to calculate the time delay between the measured lateral acceleration a y and the lateral acceleration estimated by UKF , and the specific process is as follows:

[0076] First, define the cross-correlation function between the measured value and the estimated value of the lateral acceleration:

[0077]

[0078] where R(τ) is the cross-correlation function of the two signals; T is the current time; N is the time window length of the two signals. At the maximum value of the final cross-correlation function, the delay time τ of the sequence estimation of the two signals is obtained est .

[0079] Since the delay of the lateral acceleration response is caused by the lateral movement of the tire, and at this time the estimated value of the lateral acceleration calculated based on the steady-state model will be ahead of τ est compared with the measured value. Therefore, when calculating the lateral slip ratio in the standard UKF, the delay time τ needs to be added est , so as to ensure that the measured value of the lateral acceleration and the estimated value of the lateral acceleration are on the same time sequence. Therefore, the delay time τ is added to the lateral slip ratio in Step 1 est , that is:

[0080]

[0081]

[0082] where s yfτ is the lateral slip ratio of the front axle after time delay, and s yrτFor the lateral slip ratio of the rear axle after time delay, the two slip ratios in the above formula are incorporated into the Burckhardt model (vehicle three-degree-of-freedom model) to achieve compensation for the transient characteristics of the tire, thereby ensuring that the estimated lateral acceleration and the measured value of the lateral acceleration have the same time sequence, and improving the estimation accuracy of the road surface adhesion coefficient. For the specific improvement effect, please refer to the simulation verification in Step Five.

[0083] Step Five: Software-in-the-loop Simulation Verification

[0084] According to the method proposed above, verification was carried out in the co-simulation environment of Carsim-Simulink.

[0085] Calibration value ∈ = 0.6 m / s 2

[0086] The test results are as Figure 6 、 Figure 7 、 Figure 8 、 Figure 9 shown.

[0087] In Figure 6 and Figure 7 , the estimated value of Scheme One represents the estimated value of the standard UKF method; the estimated value of Scheme Two represents adding the judgment of the lateral acceleration estimation error on the basis of the standard UKF method, but the influence of the delay time is not considered in the lateral acceleration estimation error; the estimated value of Scheme Three represents adding the judgment of the lateral acceleration estimation error, the estimation and compensation of the delay time based on UKF, that is, the method proposed in the present invention.

[0088] From Figure 6 it can be seen that the estimated value of the road surface adhesion coefficient in Scheme One cannot quickly and accurately track when the road surface adhesion coefficient changes suddenly. In Scheme Two, since the covariance matrix P is reset, the standard UKF algorithm eliminates the influence of historical data on the covariance matrix, but the oscillation degree of the estimated value of this scheme is greater than that of Scheme Three (the method of the present invention). It can be found that this method can quickly track the true value of the road surface adhesion coefficient when the road surface adhesion coefficient changes suddenly and ensure a small static error.

[0089] From Figure 7 it can be seen that there is a large error in the estimated value of the lateral acceleration in Scheme One and it is ahead of the true value of the lateral acceleration. The error between the estimated value of the lateral acceleration and the true value in Scheme Two is small, but there is a phase lead. The error between the estimated value of the lateral acceleration and the true value in Scheme Three (the method of the present invention) is small and the phase can be kept consistent.

[0090] Figure 8 And Figure 9 are respectively the front wheel steering angle and vehicle speed change diagrams during the simulation process, and it can be known that this simulation condition is a pure lateral condition.

[0091] The above description is only a description of the preferred embodiments of the present application and does not limit the scope of the present application in any way. Any change or modification made by any person skilled in the art based on the technical content disclosed above shall be regarded as an equivalent effective embodiment and fall within the scope of protection of the technical solution of the present application.

Claims

1. Method for estimating road surface adhesion coefficient considering tire transient characteristics under lateral conditions, comprising the following steps: Step 1: Vehicle dynamics modeling For the problem of estimating road surface adhesion coefficient under lateral conditions, a three-degree-of-freedom vehicle model is selected. The three-degree-of-freedom vehicle model includes three degrees of freedom: longitudinal motion, lateral motion, and rotation about the z-axis of the vehicle coordinate system; Step 2: Estimation of road surface adhesion coefficient based on unscented Kalman filter Based on the vehicle dynamics model established in Step 1, the relationship between the road surface adhesion coefficient and the observed variables is obtained; Taking the road surface adhesion coefficient as the variable to be estimated, the observation variable matrix includes longitudinal acceleration, lateral acceleration and yaw rate, where a x is the longitudinal acceleration, a y is the lateral acceleration, is the yaw rate; The state variables in the vehicle dynamics model include longitudinal speed, lateral speed, front and rear axle wheel speeds, and steering wheel angle, and the vehicle structure parameters include the distance from the center of mass to the front axle, wheelbase, track width, steering ratio, and effective tire radius; Step 3: Unscented Kalman filter estimation error adjustment When the slip ratio is small, the relationship between the tire force and the slip ratio is consistent under different road adhesion coefficients, and the update process of the unscented Kalman filter covariance matrix is maintained; if the lateral acceleration estimation residual exceeds the threshold ε, it is determined that the road adhesion coefficient of the current vehicle driving road has changed. At this time, the covariance matrix P corresponding to the road adhesion coefficient is reset to the initial prior value, that is, P = P0, so as to break the low covariance constraint accumulated by the filter history and significantly improve the convergence speed of the road adhesion coefficient estimation value; Step 4: Time delay estimation Calculate the time delay between the measured value and the estimated value of the lateral acceleration through the time cross-correlation function, and delay the lateral slip ratio according to the calculated time delay to achieve the time sequence calibration between the measured value and the estimated value of the lateral acceleration.

2. The method for estimating road surface adhesion coefficient considering tire transient characteristics under lateral conditions according to claim 1, in the said Step 1, vehicle model modeling: The three-degree-of-freedom vehicle model is as follows: ma x =F xr -F yf sinδ+F xf cosδ ma y =F yr +F yf cosδ+F xf sinδ Wherein, m is the mass of the vehicle; F xf , F xr are the longitudinal forces on the front axle and rear axle of the vehicle respectively; F yf , F yr are the lateral forces on the front axle and rear axle of the vehicle respectively; a x , a y are the longitudinal acceleration, lateral acceleration and yaw rate of the vehicle respectively; I z is the moment of inertia of the vehicle about the Z-axis; a and b are the distances from the center of mass to the front axle and rear axle respectively; δ is the steering angle of the front wheels of the vehicle; Then establish the formula for the longitudinal and lateral forces of the front and rear axles in the above vehicle three-degree-of-freedom model as follows: where s xi is the longitudinal slip ratio of the front and rear axles, i = {f, r}; r dyn is the effective radius of the tire; w wi is the wheel speed of the front and rear axles; v x is the longitudinal speed at the vehicle's center of mass; v y is the lateral speed at the vehicle's center of mass; is the yaw angular velocity at the vehicle's center of mass; s i is the vehicle's combined slip ratio; f(s i ) is the nominal tire force; s yf is the lateral slip ratio of the front axle; s yr is the lateral slip ratio of the rear axle; F zi is the vertical load on the front and rear of the vehicle; c1 and c2 are respectively parameters in the tire model, which are determined by the tire characteristics; μ is the road surface adhesion coefficient, which is determined by the vehicle driving environment; Where, L is the distance between the front and rear axles of the vehicle; g is the acceleration due to gravity; h is the height of the center of mass from the ground.

3. The method for estimating road surface adhesion coefficient considering tire transient characteristics under lateral conditions according to claim 1, in the said Step 3, unscented Kalman filter estimation error adjustment: When it is detected that the error between the estimated value of the lateral acceleration and the measured value a y is greater than ε, the covariance matrix P will be reset to P0 = 1×10 -5 .

4. The road surface adhesion coefficient estimation method considering tire transient characteristics under lateral conditions according to claim 1, step four: time delay estimation First, define the cross-correlation function between the measured value and the estimated value of the lateral acceleration: Among them, R(τ) is the cross-correlation function of two signals; T is the current moment; N is the time window length of the two signals. At the maximum value of the final cross-correlation function, the delay time τ of the sequence estimation of the two signals is obtained. est ; Then, add the delay time τ to the lateral slip ratio in Step 1 est , that is: where s yfτ is the lateral slip ratio of the front axle after time delay, and s yrτ is the lateral slip ratio of the rear axle after time delay. Incorporating the two slip ratios into the vehicle three-degree-of-freedom model can achieve compensation for the transient characteristics of the tires, and thus the estimated lateral acceleration and the measured value of the lateral acceleration have the same time sequence.

Images